Inverse modeling
description
Transcript of Inverse modeling
Inverse modelingSemi-analytical algorithms : sensitivity
analysis of the values of the backscattering spectral dependency n and the a_nap&cdom
slope S on two stations
Maéva DORON, Ocean Optics 2004
To begin with
• Roesler and Perry ’95 model of inverse modeling
• 3 basis vectors in input:1. spectral slope of the absorption of NAP and CDOM S (DL);
2. spectral slope of backscattering n in nm-1;3. spectral absorption of chlorophyll (normalized).
• 3 parameters in output :1. chl in mg.m-3;2. a_cdom&nap (440nm) in m-1; 3. bb(440nm) in m-1.
• Data from the second cruise: one station off the DMC dock and one station near the ocean.
Goal
• First idea• Comparison of the data retrieved by the
GSM model and the R&P95 model (hyperspectral & multispectral at 412, 442, 490, 510 and 555 nm);
• Sensitivity to S and n to the retrieved values
• Comparison of the sensitivities for the hyperspectral and the multispectral model
Original values Station A – Off of DMC Dock
a_phi S n/η [chl], ag(443) + anap(443)
bbp(443)
GSM – Maritorena et al. 2002
GSM 02 0.0206 1.0337 2.1328 0.2424 0.0203
Hyper RP CR avg. 0.0145 01
0.1827 0.3732 0.0161 0.0045
Hyper RP CR avg. 0.010.02
01
1.3146 0.2535 0.1169 0.01570.0069
RP @ GSM lambda
CR avg. 0.0145 01
1.02 0.306 0.045-0.0167
Original values Station B – SW of Pemaquid Pt.
a_phi S n/ η [chl], ag(443) + anap(443)
bbp(443)
GSM – Maritorena et al. 2002
GSM 02 0.0206 1.0337 1.5502 0.1288 0.0039
Hyper RP CR avg. 0.0145 0 1
0.8828 0.2158 0.00240.0024
Hyper RP CR avg. 0.010.02
01
0.8571 0.0899 0.1352 0.00280.0024
RP @GSM lambda
CR avg. 0.0145 01
1.00 0.187 0.00320.0019
Sensitivity
• One spectral dependency n ,instead of 0 for large particles and 1 for small particles.
• Varying the n – values : from 0 to 2.
• Change the values of S : from 0.005 to 0.04 nm-1 :
Realistic range of Scdom = 0.007–0.013nm-1
Realistic range of Snap = 0.016–0.022nm-1
Sensitivity for the n(bb) slope - hyperspectral
Sensitivity for the S(cdom&nap) slope - hyperspectral
Sensitivity for the n(bb) dependency - multispectral
Sensitivity for the S(cdom&nap) slope - multispectral
Sensitivity for the S(cdom&nap) slope – multispectral- different scales
Discussion
• This was a first qualitative approach.
• The results are encouraging because for realistic values of S are not erratic.
• Any idea about how to test the robustness of the algorithm with multispectral data?